DEV Community

Serguey Shinder
Serguey Shinder

Posted on

Ask the Business Which Days It Cannot Lose

We state availability as a percentage of a year, which is a convenient unit and a slightly dishonest one. Ninety-nine point five percent sounds exacting until you notice it permits forty-four hours of downtime, and says nothing at all about when those hours fall.

The company does not experience the year evenly. Ours has about six dates that genuinely matter. The last two working days of each quarter, when the sales team books everything it has been holding. The payroll run. The annual regulatory submission, which has a statutory deadline and no tolerance in it. And the two days after the seasonal catalogue lands, when order volume runs at four times the ordinary level. Everything else is a normal Tuesday, and a normal Tuesday can absorb an hour of disruption without anyone outside the department noticing.

I learned this by apologising for the wrong outage. We had a bad month with two incidents. The longer one, three hours on a Thursday in February, generated almost no comment. The shorter one, forty minutes on the thirtieth of September, was raised at board level, because a quarter's revenue recognition depended on transactions that could not be entered. Same estate, same cause, entirely different events.

Our planning had none of this in it. Load testing was done against average volumes rather than the catalogue peak. On-call was staffed identically in every week of the year. Our change calendar avoided Christmas, which is an IT tradition rather than a fact about this business, and cheerfully allowed work on the last day of a quarter. Worse, two of our suppliers scheduled their own maintenance into windows that collided with our close, because nobody had ever told them.

So we asked. Every department named the dates and hours it cannot lose, with a reason. The combined calendar has nineteen entries, two of which had never been mentioned to IT in a decade. Changes near those dates need the process owner's agreement. Capacity is proven at peak volume, not average. Cover is staffed against the calendar. Supplier maintenance is negotiated around it.

Availability is not really a property of a year. It is a property of particular mornings, and the business already knows which ones they are.

– Serguey Shinder

Top comments (0)